ARTEX Autonomous AI Pen-Testing Platform Abused for Large-Scale Agentic Cyberattacks
Check Point Research (CPR) uncovered widespread criminal use of ARTEX, an AI-powered penetration testing platform that connects LLMs with autonomous agents capable of planning, exploiting and documenting attacks with minimal human direction. CPR identified 2,557 systems hosting ARTEX, including 1,608 live, publicly accessible instances, and found 10 misconfigured instances exposing their full operator databases, revealing models used, operator instructions and agent actions. Campaigns targeted cryptocurrency/digital asset services, online gambling platforms, payment infrastructure, OTC crypto trading/custody services, and gray-market iOS signing services, exploiting weak/hardcoded passwords, missing authentication, IDOR and RCE flaws. Despite achieving serious technical compromises, autonomy has limits: a crypto-theft campaign generating 400,791 agent messages across 292 objectives yielded $0 profit, with agents misreporting success due to misunderstanding API responses.
Detalhes técnicos
Operators largely set initial goals and let agent fleets work autonomously, checking progress rather than directing each step. Across 10 exposed instances, Chinese AI models (notably DeepSeek Flash) dominated; American models (including Anthropic's Opus 5.5) were used on three instances but triggered guardrails, prompting operators to switch to more permissive Chinese models. In one operation, agents independently searched compromised gray-market iOS signing shops, validating stolen Apple code-signing certificates via passwords, private keys, revocation status and provisioning profiles, obtaining over 300 certificates (three fully validated, usable to sign arbitrary iOS apps) at an estimated AI cost of $700–$1,500 for ~36 billion tokens. In a second campaign, an agent tasked with stealing cryptocurrency discovered an unauthenticated API exposing AWS S3 credentials, gained write access to a bucket hosting iOS apps and landing pages, and independently executed and verified a supply chain compromise (file later restored, no confirmed profit). ARTEX C&C infrastructure was hosted mainly in the United States (54%), Hong Kong (18%) and China (6%). ARTEX was originally open-source for authorized security testing but went closed-source after being observed in attacks on South Korean banks; maintained forks keep it publicly available.
Remediação
Organizations should expect increased automated attack pressure against internet-facing systems and continuously test and harden exposed infrastructure, focusing on weak authentication, exposed APIs, credential leakage, RCE vulnerabilities and cloud storage permission misconfigurations.
Linha do tempo
-
ARTEX Autonomous AI Pen-Testing Platform Abused for Large-Scale Agentic Cyberattacks
ler a análise desta fonte ocultar
Check Point Research (CPR) uncovered widespread criminal use of ARTEX, an AI-powered penetration testing platform that connects LLMs with autonomous agents capable of planning, exploiting and documenting attacks with minimal human direction. CPR identified 2,557 systems hosting ARTEX, including 1,608 live, publicly accessible instances, and found 10 misconfigured instances exposing their full operator databases, revealing models used, operator instructions and agent actions. Campaigns targeted cryptocurrency/digital asset services, online gambling platforms, payment infrastructure, OTC crypto trading/custody services, and gray-market iOS signing services, exploiting weak/hardcoded passwords, missing authentication, IDOR and RCE flaws. Despite achieving serious technical compromises, autonomy has limits: a crypto-theft campaign generating 400,791 agent messages across 292 objectives yielded $0 profit, with agents misreporting success due to misunderstanding API responses.
Technical Details
Operators largely set initial goals and let agent fleets work autonomously, checking progress rather than directing each step. Across 10 exposed instances, Chinese AI models (notably DeepSeek Flash) dominated; American models (including Anthropic's Opus 5.5) were used on three instances but triggered guardrails, prompting operators to switch to more permissive Chinese models. In one operation, agents independently searched compromised gray-market iOS signing shops, validating stolen Apple code-signing certificates via passwords, private keys, revocation status and provisioning profiles, obtaining over 300 certificates (three fully validated, usable to sign arbitrary iOS apps) at an estimated AI cost of $700–$1,500 for ~36 billion tokens. In a second campaign, an agent tasked with stealing cryptocurrency discovered an unauthenticated API exposing AWS S3 credentials, gained write access to a bucket hosting iOS apps and landing pages, and independently executed and verified a supply chain compromise (file later restored, no confirmed profit). ARTEX C&C infrastructure was hosted mainly in the United States (54%), Hong Kong (18%) and China (6%). ARTEX was originally open-source for authorized security testing but went closed-source after being observed in attacks on South Korean banks; maintained forks keep it publicly available.
Remediation
Organizations should expect increased automated attack pressure against internet-facing systems and continuously test and harden exposed infrastructure, focusing on weak authentication, exposed APIs, credential leakage, RCE vulnerabilities and cloud storage permission misconfigurations.